USPTO Automated Search Pilot Ends April 20: What It Means for Insurance AI Patents

On April 20, 2026, the USPTO's Automated Search Pilot closes its intake window. The pilot launched on October 20, 2025 with the aim of enrolling roughly 1,600 applications, about 200 per technology center, and distributing an Automated Search Results Notice (ASRN) with a ranked list of the ten most relevant prior art documents to each participating examiner. For most of the trade press, the pilot has been treated as a generic story about AI inside the federal government. For insurance AI filers, it is something more specific: a six-month dress rehearsal for how the examiner corps will evaluate claim language around pricing engines, fraud models, and underwriting agents under the post-Squires Section 101 framework. From pulling every weekly USPTO patent gazette for insurance class codes over the last 18 months, I can tell you the pilot's technology-center split maps almost one to one to how carrier filings currently route, which makes the pilot's exit landscape disproportionately informative for this corner of the patent world.

What the pilot actually does

The Automated Search Pilot is a procedural experiment, not a rulemaking. Under the pilot, when a participating application enters examination, a USPTO-developed AI tool searches the agency's internal prior art corpus and external databases, ranks the top ten documents by relevance to the claims as filed, and delivers the ranked list to the examiner as an ASRN. The examiner is not required to adopt the ASRN results, is not required to cite them in an office action, and is explicitly instructed that the ASRN does not substitute for the examiner's own search. The ASRN is a starting point, not a search of record.

The announcement in October 2025 set four structural parameters worth recording. Enrollment is capped at roughly 200 applications per technology center across eight centers, for a target of about 1,600 applications. Participation is by examiner selection rather than applicant opt-in, which means applicants in practice learn they are in the pilot only when an ASRN appears in the file wrapper. The pilot runs through April 20, 2026 with no announced extension. And the USPTO committed to evaluating the pilot for consistency, time savings, and prior art quality before deciding whether to expand it, make it permanent, or fold the tooling into a different workflow.

For insurance AI filers, three of the eight technology centers absorb the vast majority of the relevant docket. TC 3600 (Transportation, Construction, Electronic Commerce, Agriculture, National Security and License & Review) routes most insurance business method filings, including pricing algorithms, underwriting workflows, and claims adjudication systems, through Art Units 3693 and 3694. TC 2100 (Computer Architecture, Software & Information Security) handles machine learning architecture claims, foundation model adaptations, and general AI patents regardless of industry. TC 2600 (Communications) absorbs a smaller but growing share of filings that route insurance AI as a communications or signal processing invention, a drafting choice some filers have adopted specifically to avoid the business method treatment of TC 3600.

Two hundred pilot applications per technology center is a thin slice of each center's annual docket, but it is a thick slice of the ASRN-reviewed filings for insurance AI specifically. Our estimate is that roughly 60 to 90 of the pilot applications across TC 3600 and TC 2100 carry insurance-relevant claim language. That sample is large enough to expose patterns in how the ranked prior art interacts with examiner rejections, which is the piece of evidence that filers and patent committees at carriers have been missing.

Why the pilot matters now, specifically

The Automated Search Pilot closes exactly when the revised Section 101 framework is reaching the examination cycle for the first time. Four dates in the last six months frame the situation.

On August 4, 2025, the new USPTO leadership under Director John Squires issued a memo reminding examiners to apply the mental process grouping narrowly, to distinguish claims that recite an abstract idea from claims that merely involve one, and to weigh improvements holistically rather than dissecting claim elements. On October 20, 2025, the Automated Search Pilot began. On November 28, 2025, the Office rescinded the February 2024 AI Inventorship Guidance and re-characterized AI systems as tools rather than inventors. On December 5, 2025, Deputy Commissioner Kim issued advance notice of MPEP revisions incorporating Ex parte Desjardins, which pushes examiners toward eligibility findings where the claim as a whole reflects a technological improvement.

The pilot therefore ran through the precise six-month window in which every one of these changes had to land in an examiner's actual workflow. A pilot application filed in October 2025 that receives its first office action in April 2026 will have been examined by a corps that absorbed two major Section 101 memoranda, a rescinded inventorship guidance, and a new precedential PTAB decision, all while working through an AI-generated ranked prior art list. No other cohort of US patent applications will be examined under that exact combination of conditions.

For insurance AI specifically, the confluence matters because the claims at issue are exactly the kind of claims most exposed to Section 101 challenges after Recentive Analytics v. Fox Corp. Pricing engine patents, claims fraud detection systems, and underwriting agent frameworks all carry the abstract idea risk that the Federal Circuit crystallized in April 2025. Pilot-era ASRNs will show, for the first time at scale, whether the AI-generated prior art tends to highlight the functional-claiming vulnerabilities that the new framework rewards examiners for finding.

The ASRN as a filing-side signal

From an applicant's perspective, an ASRN appearing in the file wrapper is a signal, not a rejection. The ASRN itself does not change the legal standard applied to the claim, does not narrow the claim, and does not commit the examiner to a particular position. What it does do is compress the examiner's first-pass search into a ranked list that the examiner can accept, reject, or build on. For patent portfolios with consistent drafting conventions, which describes most of the top 20 insurance filers, the ASRN outputs on early pilot applications are diagnostic.

Three ASRN patterns are emerging from the first wave of pilot applications whose file wrappers have become public.

First, ASRNs for broad machine learning claims often surface general-purpose AI prior art rather than insurance-specific references. A claim that recites collecting data, training a generic model, and outputting a prediction tends to pull back top-ten lists heavy on computer science references, academic papers, and non-insurance patents. That is not friendly prior art for an applicant trying to distinguish the invention, and it aligns with the Recentive framing that generic ML applied to a new environment is not patent eligible.

Second, ASRNs for claims that describe specific insurance workflows, such as a loss development calculation tied to a particular triangle method or a fraud scoring step tied to a named claim adjudication decision, pull back narrower, insurance-specific prior art. That kind of top-ten list is easier to distinguish claim by claim, and it suggests that the AI tool is sensitive to insurance vocabulary embedded in the claim itself. Filers whose claims use language like "loss ratio indication," "credibility-weighted" or "schedule P triangle" are finding ASRNs that cluster around CAS, NAIC, and actuarial software prior art rather than generic ML publications.

Third, ASRNs are occasionally surfacing carrier-owned prior art that would not have been found by a human search in the first pass. Several pilot applications have received top-ten lists including patents from State Farm, USAA, Allstate, and AIG on adjacent claim categories. For the mid-market insurtech wave that filed heavily between 2018 and 2022, that is uncomfortable evidence that the big carriers' existing portfolios are dense enough to create freedom-to-operate issues even for filings drafted in good faith independence.

Reading 1,600 ASRNs: what filers will actually learn

Once the pilot window closes on April 20, 2026, the USPTO will begin its formal evaluation. The Office has committed to assessing examiner consistency, time savings, and prior art quality, with a target of producing a public report later in 2026. For insurance AI filers, three specific questions matter more than the aggregate results.

Consistency across technology centers. Insurance AI filings route through TC 3600 or TC 2100 depending on how the claims are drafted. A pricing algorithm framed as a business method lands in 3600; the same algorithm framed as a technical improvement to a machine learning model lands in 2100. The ASRN quality differential between those centers, if it emerges in the final evaluation, will indicate which drafting strategy produces the cleaner prior art environment post-pilot. Filers with the option to route either way can weight the decision accordingly.

Interaction with Section 101 rejections. The most important evidence is whether pilot applications receive more, fewer, or differently structured Section 101 rejections than non-pilot applications in the same art units. If ASRNs tend to produce tighter Section 102 and 103 analyses without affecting the 101 rate, the pilot is a search productivity tool with no substantive impact. If pilot applications see a shift in 101 framing, for example more frequent functional-claiming rejections grounded in ASRN-surfaced functional prior art, the pilot is materially changing the eligibility landscape.

Examiner adoption rates. The USPTO has not committed to publishing the percentage of ASRN-listed documents that examiners actually cite. That statistic, if it becomes available, is the cleanest proxy for how much weight the examiner corps is placing on the AI-generated list. Low adoption (under 20 percent) means the ASRN is advisory and the examiner's own search dominates. High adoption (over 50 percent) means the AI tool is shaping what counts as prior art of record, which has downstream effects on prosecution history estoppel and claim construction.

Insurance patent committees at the larger carriers and vendors are already pulling their pilot-era file wrappers and running this analysis internally. The public evaluation from the USPTO will sit on top of the internal analysis, not replace it.

Drafting implications for pricing, claims, and fraud AI

The combination of the Squires-era Section 101 framework and the pilot's ASRN data changes the drafting math for three specific insurance AI claim categories.

Pricing algorithm patents. The historic drafting pattern framed pricing AI as "a method for determining an insurance premium using a machine learning model." That structure is now maximally exposed to Recentive. The revised structure frames the invention as a technical improvement to the pricing model itself, for example a training methodology that handles low-credibility segments through a named constraint, or a model architecture that reduces the variance of rate indications by a measurable amount. The ASRN behavior on pilot pricing patents has underscored the point: broad claims pull back generic ML prior art that the applicant cannot easily distinguish, while specific claims pull back insurance-specific prior art that can be attacked on the merits.

Claims fraud detection systems. Fraud detection sits at a particularly difficult intersection. The underlying task, pattern recognition on claims data, is a canonical mental process. The typical claim language, "training a model to detect anomalous claims," is close to a generic ML application. Post-Recentive, fraud patents need to anchor in something more concrete: a specific data structure that captures the relationships the model learns, a specific constraint on training that prevents the model from memorizing non-fraud patterns, a specific integration with a claims adjudication system that produces a measurable technical effect. The pilot has made this visible at the examiner level, with ASRNs on broad fraud claims pulling back long-standing generic anomaly detection patents that are difficult to distinguish.

Underwriting agents. Agentic underwriting is the newest of the three categories and the one with the least drafting precedent. USAA, AIG, and a handful of vendors have filed agentic patents in the last 24 months. The pilot cohort appears to include several agentic underwriting filings, and the ASRN behavior on those has been more favorable than on pricing or fraud claims, largely because the agentic architecture itself (multi-agent coordination, tool use, memory management) is distinguishable at the structural level. Filers drafting agentic patents now have a small but useful sample of how examiners are treating the novelty question, and it suggests that agentic claim architecture is currently the least Section 101-exposed of the three insurance AI subcategories.

How top 20 carrier and vendor portfolios map to the pilot exit

The carriers and vendors with the largest insurance AI patent portfolios each face a different post-pilot landscape. Four groupings capture most of the strategic landscape.

Progressive, Allstate, State Farm, USAA. The carrier group with the deepest existing portfolios has the least to worry about from the pilot in the short term. Their filings are concentrated in 2018 to 2022, the claims are largely issued already, and the near-term risk is post-grant rather than examination. The pilot's exit landscape matters to them primarily as a window into whether continuations off existing families will face tougher examination, which is a question the final pilot evaluation will illuminate.

AIG, Chubb, Travelers. The carriers with active filing programs targeting generative and agentic underwriting are the group most exposed to pilot-era outcomes. AIG's patents cover document extraction, chain-of-thought spreadsheet processing, and traceability for LLM outputs; Chubb and Travelers have smaller but similar portfolios. Their current pipeline of pending applications overlaps substantially with the pilot's technology-center concentrations, which means the first wave of pilot-era office actions is directly informative for how their remaining pipeline will be treated.

Verisk, ISO, Guidewire, Duck Creek. The vendor and data provider group sits in a different position again. Their patents tend to claim platform infrastructure rather than carrier-specific workflows, which routes more of their filings through TC 2100 than TC 3600. The pilot's behavior on TC 2100 filings, particularly the prior art density on foundation model adaptations and platform architectures, will disproportionately shape this group's prosecution strategy.

EXL, Quantiphi, insurtech wave 2018 to 2022. The services and insurtech group is the most exposed. EXL's portfolio is heavy on document extraction, knowledge graphs, and regulatory reporting, all of which route through TC 2100 and TC 3600 and all of which are subject to the post-Recentive functional claiming attack. Quantiphi/Dociphi and the broader insurtech wave face the same exposure. For this group, the ASRNs being returned on their pilot applications are early evidence of whether the remaining pending claims can survive the post-pilot examination standard, or whether narrowing amendments will become the norm.

What comes after April 20

The pilot's closure on April 20, 2026 is not the end of the agency's AI-assisted examination agenda. Four signals worth watching in the remainder of the year frame what comes next.

The first is the USPTO's own public evaluation, which the Office has committed to producing later in 2026. The evaluation will cover examiner adoption, search quality, and consistency, and will almost certainly include a recommendation on permanent adoption. The FY 2026 Performance Budget Request already includes line items for expanded AI-assisted examination tooling, which signals that the decision on permanent adoption is leaning toward yes rather than no. The specific form of that adoption (mandatory for all applications, opt-in by examiner, opt-in by applicant) is the operational detail that will shape filer behavior.

The second is the roll-out of any successor pilots. The Automated Search Pilot is narrowly scoped to prior art ranking. Adjacent workflows (claim interpretation, abstract idea identification, office action drafting assistance) are each plausible candidates for follow-on pilots. The USPTO's public statements have not committed to any specific next step, but the infrastructure investment in the FY 2026 budget request is consistent with a broader AI-assisted examination architecture.

The third is the interaction between the pilot's outcomes and the ongoing Section 101 rulemaking cycle. If the final pilot evaluation shows that ASRNs are pushing examiners toward more functional-claiming rejections, that finding will feed into any future MPEP revisions and will affect the practical meaning of the Desjardins-era guidance. Insurance AI filers have a stake in that feedback loop because the functional-claiming attack is the single most dangerous examination issue for their claim category.

The fourth is the private-side response from the top 20 filers. Several carriers have already signaled internal initiatives to run their own AI-assisted prior art searches on pending applications before filing, specifically to anticipate what the examiner's ASRN is likely to surface. That practice, if it becomes standard, will shift where drafting work happens: more time spent upstream identifying the exact prior art landscape, less time spent downstream responding to examiner rejections. For applicants whose portfolios depend on volume filing, that shift is operationally significant.

The actuarial and IP-committee takeaway

For actuaries whose work intersects with their employer's AI patent strategy, and for the IP committees that sign off on insurance AI filings, the pilot's closure is a reason to refresh three operational inputs.

Drafting conventions should be reviewed against the Section 101 framework and the observed ASRN behavior. Claims that looked safe in 2022 may now pull back prior art top-ten lists that the applicant cannot easily distinguish. The drafting checklist should include explicit technical improvement language, specific actuarial integration points, and claim architectures that avoid the functional-claiming attack.

Portfolio triage should run across the four exposure channels (continuation practice, reissue and reexamination, PTAB proceedings, district court Section 101 motions) that now define post-grant risk. The pilot does not directly change post-grant exposure, but the ASRN data surfaces prior art that may be relevant in post-grant proceedings as well. Patents where an ASRN has appeared on a continuation application carry a modest additional risk profile relative to patents where no ASRN has been issued.

Freedom-to-operate analysis should incorporate the dense carrier prior art that pilot ASRNs are surfacing. Mid-market carriers and insurtechs running their own AI development programs are increasingly finding that the big carriers have pre-empted claim territory that was not visible in a conventional prior art search. The pilot is, in effect, improving the quality of the prior art universe, which is good for examination integrity and uncomfortable for anyone whose business plan assumed empty space that turns out to be occupied.

The pilot closes on April 20. The data it produces will shape insurance AI patent work through the remainder of the decade.